[CI/Build] Replaced some models on tests for smaller ones (#9570)

Signed-off-by: Wallas Santos <wallashss@ibm.com>

tests/basic_correctness/test_basic_correctness.py

@@ -19,7 +19,7 @@ from ..utils import multi_gpu_test
 
 MODELS = [
     "facebook/opt-125m",
-    "meta-llama/Llama-2-7b-hf",
+    "meta-llama/Llama-3.2-1B",
 ]
 
 TARGET_TEST_SUITE = os.environ.get("TARGET_TEST_SUITE", "L4")

tests/basic_correctness/test_chunked_prefill.py

@@ -16,7 +16,7 @@ from ..utils import multi_gpu_test
 
 MODELS = [
     "facebook/opt-125m",
-    "meta-llama/Llama-2-7b-hf",
+    "meta-llama/Llama-3.2-1B",
 ]
 
 

tests/basic_correctness/test_cpu_offload.py

@@ -2,5 +2,5 @@ from ..utils import compare_two_settings
 
 
 def test_cpu_offload():
-    compare_two_settings("meta-llama/Llama-2-7b-hf", [],
-                         ["--cpu-offload-gb", "4"])
+    compare_two_settings("meta-llama/Llama-3.2-1B", [],
+                         ["--cpu-offload-gb", "1"])

tests/compile/test_basic_correctness.py

@@ -13,8 +13,7 @@ from ..utils import compare_all_settings
 @pytest.mark.parametrize(
     "model, model_args, pp_size, tp_size, attn_backend, method, fullgraph",
     [
-        ("meta-llama/Meta-Llama-3-8B", [], 2, 2, "FLASH_ATTN", "generate",
-         True),
+        ("meta-llama/Llama-3.2-1B", [], 2, 2, "FLASH_ATTN", "generate", True),
         ("nm-testing/Meta-Llama-3-8B-Instruct-W8A8-Dyn-Per-Token-2048-Samples",
          ["--quantization", "compressed-tensors"
           ], 1, 1, "FLASH_ATTN", "generate", True),

tests/entrypoints/llm/test_chat.py

@@ -8,7 +8,7 @@ from ..openai.test_vision import TEST_IMAGE_URLS
 
 
 def test_chat():
-    llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    llm = LLM(model="meta-llama/Llama-3.2-1B-Instruct")
 
     prompt1 = "Explain the concept of entropy."
     messages = [
@@ -26,7 +26,7 @@ def test_chat():
 
 
 def test_multi_chat():
-    llm = LLM(model="meta-llama/Meta-Llama-3-8B-Instruct")
+    llm = LLM(model="meta-llama/Llama-3.2-1B-Instruct")
 
     prompt1 = "Explain the concept of entropy."
     prompt2 = "Explain what among us is."

tests/entrypoints/openai/test_chat.py

@@ -16,9 +16,6 @@ from .test_completion import zephyr_lora_files  # noqa: F401
 
 # any model with a chat template should work here
 MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
-# technically this needs Mistral-7B-v0.1 as base, but we're not testing
-# generation quality here
-LORA_NAME = "typeof/zephyr-7b-beta-lora"
 
 
 @pytest.fixture(scope="module")

tests/entrypoints/openai/test_shutdown.py

@@ -6,7 +6,7 @@ import pytest
 
 from ...utils import RemoteOpenAIServer
 
-MODEL_NAME = "HuggingFaceH4/zephyr-7b-beta"
+MODEL_NAME = "meta-llama/Llama-3.2-1B"
 
 
 @pytest.mark.asyncio

tests/test_sharded_state_loader.py

@@ -46,9 +46,10 @@ def test_filter_subtensors():
 @pytest.fixture(scope="module")
 def llama_2_7b_files():
     with TemporaryDirectory() as cache_dir:
-        input_dir = snapshot_download("meta-llama/Llama-2-7b-hf",
+        input_dir = snapshot_download("meta-llama/Llama-3.2-1B",
                                       cache_dir=cache_dir,
-                                      ignore_patterns="*.bin*")
+                                      ignore_patterns=["*.bin*", "original/*"])
+
         yield input_dir
 
 
@@ -58,9 +59,12 @@ def _run_writer(input_dir, output_dir, weights_patterns, **kwargs):
     # Dump worker states to output directory
     llm_sharded_writer.llm_engine.model_executor.save_sharded_state(
         path=output_dir)
+
     # Copy metadata files to output directory
     for file in os.listdir(input_dir):
-        if not any(file.endswith(ext) for ext in weights_patterns):
+        if not any(
+                file.endswith(ext) and not os.path.isdir(file)
+                for ext in weights_patterns):
             shutil.copy(f"{input_dir}/{file}", output_dir)